Bases: Module
The decoder postnet that projects the model output vectors into token
predictions.
Source code in speechain/module/postnet/token.py
| class TokenPostnet(Module):
"""The decoder postnet that projects the model output vectors into token
predictions."""
def module_init(self, vocab_size: int, input_dim: int = None):
"""
Args:
input_dim: int
The dimension of the output vectors from the decoder
vocab_size: int
The number of tokens in the dictionary.
"""
# input_size and output_size initialization
if self.input_size is not None:
input_dim = self.input_size
else:
assert input_dim is not None
self.output_size = vocab_size
# para recording
self.input_dim = input_dim
self.vocab_size = vocab_size
# initialize the output layer
self.linear = torch.nn.Linear(in_features=input_dim, out_features=vocab_size)
def forward(self, input: torch.Tensor):
"""
Args:
input:
Returns:
"""
return self.linear(input)
|
forward(input)
Parameters:
| Name |
Type |
Description |
Default |
input
|
Tensor
|
|
required
|
Returns:
Source code in speechain/module/postnet/token.py
| def forward(self, input: torch.Tensor):
"""
Args:
input:
Returns:
"""
return self.linear(input)
|
module_init(vocab_size, input_dim=None)
Parameters:
| Name |
Type |
Description |
Default |
input_dim
|
int
|
int
The dimension of the output vectors from the decoder
|
None
|
vocab_size
|
int
|
int
The number of tokens in the dictionary.
|
required
|
Source code in speechain/module/postnet/token.py
| def module_init(self, vocab_size: int, input_dim: int = None):
"""
Args:
input_dim: int
The dimension of the output vectors from the decoder
vocab_size: int
The number of tokens in the dictionary.
"""
# input_size and output_size initialization
if self.input_size is not None:
input_dim = self.input_size
else:
assert input_dim is not None
self.output_size = vocab_size
# para recording
self.input_dim = input_dim
self.vocab_size = vocab_size
# initialize the output layer
self.linear = torch.nn.Linear(in_features=input_dim, out_features=vocab_size)
|